Hello, thanks for the work.
I cloned the code and run it on my PC. Since my graphics card is GTX1660S, I have to reduce 'train_num_rays' to 128 in case OOM error occurs(other parameters remain default). Then I extract the mesh and found it has some noise and holes. I thought maybe it's hard for my card to train a good result:(
So, would you mind share some pretrained models of, for example, nerf synthetic datasets? Or share some ideas of tuning those parameters so that a 6GB card can get good result? Thanks!
Hi! You should also try to reduce max_train_num_rays and ray_chunk. Note that you may need more training iterations as the number of sampled rays per iteration is lower.
Hello, thanks for the work. I cloned the code and run it on my PC. Since my graphics card is GTX1660S, I have to reduce 'train_num_rays' to 128 in case OOM error occurs(other parameters remain default). Then I extract the mesh and found it has some noise and holes. I thought maybe it's hard for my card to train a good result:( So, would you mind share some pretrained models of, for example, nerf synthetic datasets? Or share some ideas of tuning those parameters so that a 6GB card can get good result? Thanks!